Claude Code Receipts

receipts

[](https://github.com/monikagadage/claude-code-receipts/actions/workflows/test.yml) [](LICENSE) 

**Claude Code says "Done ✅". This checks whether it actually did the thing.**

`receipts` is a Claude Code plugin that runs when Claude finishes a turn and compares what it *claimed* against what the session transcript *shows*:

• "all tests pass" — but no test command ran this session
• "I updated src/auth.py" — but nothing touched that file
• "as we agreed, switching to GraphQL" — but you never said that

It's a `Stop` hook. Deterministic — no network calls, no model, no API cost. It reads the transcript Claude Code already writes to disk. **Warn-only by default**: it surfaces the unbacked claims and lets you decide.

⚠️  receipts — 2 claims not backed by this session:
  • "All tests pass."          → no test command ran this session
  • "I refactored auth.py."    → nothing modified `auth.py`  (files edited: routes.py, config.py)

Why

"Claude claimed the work was done and it wasn't" is one of the most common Claude Code complaints of 2026. The usual advice is a habit ("always ask for receipts") or a Stop hook that runs a *fixed* command like `npm test`. `receipts` is different: it reads the **specific claims** in Claude's final message and checks each one, with zero configuration.

Install

/plugin marketplace add monikagadage/claude-code-receipts
/plugin install receipts@monikagadage

Requires `python3` on your `PATH` (3.9+). No pip packages.

To try it without installing:

git clone https://github.com/monikagadage/claude-code-receipts
claude --plugin-dir ./claude-code-receipts

What it checks (v1)

Claim in Claude's final message	Backed by	Flagged when
"tests